Machine learning in trading: theory, models, practice and algo-trading - page 1461

 
elibrarius:

Yes - the difference between the current price and those 50 bars.
I haven't tried all the easier ones yet. I'll postpone the trends for the future.

The simple one goes straight to the landfill - nonsense. On the machine I start a bunch of indicators, I get a bunch of different models, even a million, in different combinations. This is bullshit.

It's easier to make sense of what time series memory is, wrote above. Then Delusional ideas fall away without losing time
 
Maxim Dmitrievsky:

The simple one goes straight to the landfill - nonsense. On the machine I start a bunch of indicators, I get a bunch of different models, even a million, in different combinations. This is bullshit.

I mean, you used to be a supporter of the pure price. And you think that any indicator may be built inside the NS if necessary.
Of course, it is not able to add and multiply and it cannot create indicators inside itself, but it memorizes templates from price or indicator charts very well.
 
elibrarius:
Well, you used to be the supporter of pure price work too. And that any indicator, if necessary, will be built inside the NS.
Of course, the forest is not able to add and multiply, and will not create indicators inside itself, but it is good at memorizing patterns from price charts.

Asayulenko came up with the idea that any NS indicator is built. I was just using returns - there is no information in them.

On pure prices it is possible to train on flat chunks, for trend chunks it will be out of limits of the training sample, for this it is necessary to add trends (i.e. subtract them from the price).

And yes, the bare price has the most alpha
 
Is it possible to find a platform somewhere online to open a demo account?
 
Maxim Dmitrievsky:

Asayulenko came up with the idea that any indicator is based on price

I agree with that because I tried to build a graph of digital filters from price with a neural network and it turned out very well.

Maxim Dmitrievsky:

I agree. I tried to build a digital filter graph for prices with a neural network and it worked very well.

if there has ever been such a situation in history, its development is already known. The only way out would be if such a situation has never happened, but that is extremely rare, (the moves that happened during the Brexit). Everything else has already happened many times, we just need to find the most common outcome of the event.

 
Maxim Dmitrievsky:

The fact that any NS indicator is built was invented by Asayulenko.

The level of kindergarten, including the mutilation of surnames. You should see a psychiatrist for sure.
 
elibrarius:

I agree with this, because I tried to build a graph of digital filters from the price with a neural network - it turned out very well.

If there has ever been such a situation in history, its development is already known. It will only go out of bounds if it has never happened before, but that is extremely rare (the moves that happened during the Brexit). Everything else has already happened many times, you just have to find the most frequent outcome of the event.

This is nonsense.

Out of the range of price values, do I make myself clear? the forest will only buy or sell in such a case
 
Yuriy Asaulenko:
The level of kindergarten, including the mutilation of surnames. You should definitely see a psychiatrist.

And don't peek, we'll figure it out without you. Go teach your blots, graalewood.

 
Maxim Dmitrievsky:

It's just nonsense.

Out of the range of price values, do I make myself clear? The forest will only buy or sell in this case

The TF is not nonsense, I checked it.

I use the difference, not the prices. Only very strong movements rarely occur in the history. But they are filtered out. I have fixed 1% of the strongest moves. It's more reliable to wait out such movements.)

 
Elibrarius:

TF is not nonsense, I checked it.

Well, I do not use the prices themselves, but the difference. Only very strong deficits were rare in the history. But they are filtered out and I have fixed 1% of the strongest movements. It's safer to wait out such movements.)

What does the TF have to do with it? We don't work with radio signals. Of course they don't work on fin.